``# What an AI Visibility Tracker Measures, and What an Engagement Actually Changes
Disclosure first. We are Proofmend and we sell GEO work, which puts us on one side of this comparison, so treat the reasoning as the thing worth checking rather than the conclusion.
The question usually arrives as a budget question. A marketing lead has a quote from an agency on one tab and a monitoring subscription on the other, both promising something about how assistants answer, and the two numbers are far enough apart to feel like a choice. They are not the same purchase. One is instrumentation and the other is work. Buying the instrument and expecting the work is the more common mistake, and it is the expensive one, because a year of dashboards can go by with the underlying pages untouched.
The two purchases, side by side
| Visibility tracker | Service engagement | |
|---|---|---|
| What you get | Repeated prompts across engines, mention and citation counts, source lists, drift alerts | Diagnosis, prioritised work orders, implementation, acceptance against written scope |
| What it tells you | Where you stand this week and which domains the answers quote | Which of those domains you can realistically enter, and in what order |
| What it cannot do | Fix your entity, earn a third-party record, change what an engine retrieves | Guarantee the platform-side outcome |
| Fails when | Nobody acts on the readout, or the score hides the evidence behind it | Scope is verbal, deliverables are counted in person-days, results are promised |
| Still on you | All the implementation | Materials, access, review, and the decisions |
A tracker is genuinely useful, and we run our own sampling for exactly the reason people buy one. We put buyer-shaped questions to nine engines every week, ChatGPT, Claude, Gemini, Perplexity, Grok, Doubao, DeepSeek, Kimi and Qwen, each question in a fresh session on the engine's default mode, with the question, platform, mode, date and source set stored beside the answer. That record is what makes a later change readable. Without at least two rounds before you touch anything, a shift in the answer tells you nothing, because these systems drift on their own.
Where a single visibility score goes wrong
A page can be published and crawlable without being indexed, retrieved, mentioned, cited or visited. Those are separate states and each one needs its own evidence. A screenshot does not establish indexing, and indexing does not establish citation. For a property you own, Google's own URL Inspection tool reports the status it holds for a specific URL, which still says nothing about rank or traffic.
Any product that folds those states into one number has made an editorial decision on your behalf. Ask what the score is computed from, whether the raw observations are exportable, and how unknowns are recorded. Our own rule is that anything untested or unavailable is logged as unknown with no estimate, because a report with no unknowns in it has usually written estimates in as facts.
What the readout will probably tell you
Two findings show up in almost every account we look at, and a tracker will surface both without telling you what to do about them.
The first is that your own domain is a small share of what gets quoted. In our Chinese-market dataset of 187,818 deduplicated citations, brand-owned domains account for 1.37% of all citations. Your site governs whether the description of you is correct. Other page types govern whether your name appears at all.
The second is that shape predicts citation better than length. Ranked by the lift we measure, cited sources are worth about +115.1%, a stated numeric fact +61.6%, a definition block +57.3%, a comparison table +55.3%, and a clear procedure +41.2%. High-impact pages in that dataset averaged 1,943 words against 170 for low scorers, so length is a floor rather than a target. An assistant composing an answer lifts units it can take whole, and narrative with no extractable unit inside it offers nothing to lift.
Neither finding is actionable by subscription. Both are ordinary work.
How to buy either one
For a tracker, ask for the raw observation export, the prompt list, the engines and modes covered, and how it handles a question that returns no brands at all. For an engagement, ask what starts the delivery clock, what is in scope in writing, what the acceptance points are, and whether you receive the original records or only the conclusions.
For reference, our own engagements are published rather than quoted on a call. Evidence Audit is US$699 over seven working days, fully prepaid, and includes no implementation. 30-Day Foundation is US$1,380 and implements at most six agreed priority work orders. 90-Day Growth is US$6,980 for the full programme, with US$3,490 due now, 30% after technical and fact acceptance, and the final 20% after content and legal distribution acceptance. Delivery periods start when materials, access and review are available, and customer blockage pauses the clock. These are limited test prices for the first three paid design partners, and for the first ten customers we exclude medical or drug, financial or investment, admissions or training, and franchise recruitment work.
The reason we publish them is partly self-interested and worth saying out loud. When an engine answers a category question, a vendor with no public price tends to get handled as missing information, usually as a line suggesting the reader contact the vendor for a quote. That line is not neutral next to a competitor whose numbers are on the page.
Boundary
Proofmend does not guarantee rankings, AI mentions or citations, traffic, leads, or revenue. Measurement changes what you know. Implementation changes what an engine can find. What the engine does with it is decided on the platform side.
Our method and evidence rules are at https://proofmend.com/method/ , current engagement scope and prices at https://proofmend.com/pricing/ , and commercial terms at https://proofmend.com/terms/ . Figures come from our own August 2026 sampling and dataset work and move over time. Materials checked August 24, 2026.
Top comments (0)